Color cast on print scan with a profiled Epson V800

[eriksatie]

Hi everyone, this is my first post here, I write from Italy

It's about one year that I've bought an Epson V800 to scan my personal and family photo archive. I've scanned about 1500 35mm color negatives so far with Silverfast, also thanks to Mark D Segal book. I've managed to correct the frequent color casts and I'm happy with the results.

Now I'm moving to prints and slides, and for these I've bought the IT8 targets from LaserSoft Imaging to do automatic calibration with Silverfast.

I've done the calibration but I'm unsure of the results. Silverfast says that the delta E is 0,80 for the reflective profile, that should be good enough, but the grey patches are not very neutral. grey patch n.10 measure L: 42, A: 00, B: 03 (apart from color shift, shouldn't it be near L: 50?)

My monitor is calibrated to D65 with i1display pro. The monitor is a medium range LG for office/generic use, the delta E is 0,42 so it shouldn't be so bad. I've also checked the scans with the monitor of my MacBook Pro (also calibrated) that match quite closely the LG.

Moreover, print scans have a blu/green cast that I'm able to correct with Silverfast Global CC, using the Cyan/red axis. This is non coherent with the scan of the target that shows a slight yellow tint.

But I have some prints made with Glossy Fujicolor Crystal Archive (same paper as the IT8 target) that match quite closely the scan.

So what's the problem? is the calibration "paper dependent"? Do different papers interacts differently with the scanner light? or the problem is just how I see the scans with my monitor? maybe I should calibrate the monitor to D50 for a closer match?

Maybe I was expecting to much from calibration, but I hoped to be able to have a nearly correct scan with no needs of major color correction after calibration.

I'm attaching the IT8 scanned with the custom profile applied. every input is welcome

[digitaldog]

I've done the calibration but I'm unsure of the results. Silverfast says that the delta E is 0,80 for the reflective profile, that should be good enough, but the grey patches are not very neutral. grey patch n.10 measure L: 42, A: 00, B: 03 (apart from color shift, shouldn't it be near L: 50?)

The report sounds fine, I wouldn’t worry about the Lstar patch not being 50 as the target may not actually be at this value. Best to start testing the profile with actual scans.

[smthopr]

Hi everyone, this is my first post here, I write from Italy

It's about one year that I've bought an Epson V800 to scan my personal and family photo archive. I've scanned about 1500 35mm color negatives so far with Silverfast, also thanks to Mark D Segal book. I've managed to correct the frequent color casts and I'm happy with the results.

Now I'm moving to prints and slides, and for these I've bought the IT8 targets from LaserSoft Imaging to do automatic calibration with Silverfast.

I've done the calibration but I'm unsure of the results. Silverfast says that the delta E is 0,80 for the reflective profile, that should be good enough, but the grey patches are not very neutral. grey patch n.10 measure L: 42, A: 00, B: 03 (apart from color shift, shouldn't it be near L: 50?)

My monitor is calibrated to D65 with i1display pro. The monitor is a medium range LG for office/generic use, the delta E is 0,42 so it shouldn't be so bad. I've also checked the scans with the monitor of my MacBook Pro (also calibrated) that match quite closely the LG.

Moreover, print scans have a blu/green cast that I'm able to correct with Silverfast Global CC, using the Cyan/red axis. This is non coherent with the scan of the target that shows a slight yellow tint.

But I have some prints made with Glossy Fujicolor Crystal Archive (same paper as the IT8 target) that match quite closely the scan.

So what's the problem? is the calibration "paper dependent"? Do different papers interacts differently with the scanner light? or the problem is just how I see the scans with my monitor? maybe I should calibrate the monitor to D50 for a closer match?

Maybe I was expecting to much from calibration, but I hoped to be able to have a nearly correct scan with no needs of major color correction after calibration.

I'm attaching the IT8 scanned with the custom profile applied. every input is welcome

The it-8 that you posted looks ok to my eye on my display.  The white patch is slightly blue, but the other grey patches vary from slightly yellow to slightly red.  It's quite possible that this is close to the true colors of the it-8 print.  I assume that you used the data profile for the it-8 to make your scanner profile?

So, it could be that all is fine with this scanner profile.  I think expecting perfectly colored scans for every scan might be expecting too much, but you should be in the ball park.

[eriksatie]

The it-8 that you posted looks ok to my eye on my display.  The white patch is slightly blue, but the other grey patches vary from slightly yellow to slightly red.  It's quite possible that this is close to the true colors of the it-8 print.  I assume that you used the data profile for the it-8 to make your scanner profile?

Yes Silvefast does automatic profiling using the data profile provided. the IT8 target is not manually measured, only batch measured

[eriksatie]

I'm doing some more testing and what I see in real scans is a yellow tint. I'm posting an example.
The first image has no correction, just the custom profile applied and then converted in Adobe RGB
The second image is corrected with the aim to eye match the print. correction is made inside Silverfast with Global CC. I'm attaching also the parameters used.

The print is made on AGFA Prestige Paper, while the IT8 target is made on Glossy Fujicolor Crystal Archive (if this could matter)

[eriksatie]

Another test (same paper)

The first image has no correction, just the custom profile applied and then converted in Adobe RGB
The second image is corrected with the aim to eye match the print. correction is made inside Silverfast with Global CC. Correction is very similar to previous image. Basically the scans seems to have too much yellow and lacks a bit of magenta

[saiguy]

My workflow in SF8. Give this a try. I find the Global CC too difficult to work with.

Pre scan and set the frame. Hit the ACC, auto color correction. This will set a very good white point unless it keys on some specular highlight. If that happens I will increase the brightness with the top slider in Picture Settings. For sun set shots the Landscape choice seems to work best.

Open the Histogram tool and use the Separate tab. Undock the tool and you can drag the bottom corner to stretch it across the screen. Then for each RGB channel set a good black point by moving the left slider just a little inside where the data starts.

Open the Densitometer tool. By shift clicking you can set and remove colors samplers, up to 4. Try to find what should be the darkest black place in the photo. If the RGB numbers are very close you have a neutral black point. If not, undock the Histogram again and you can fine tune the numbers on the sampler.

If you see a color cast, the Neutral Pipet tool can remove it if you find the right place to click. But you need to set in Preferences for the tool to be used for that, otherwise the place you click will go R128, G128, B128. Some times I set a color sampler before clicking with the Neutral Pipet. This can be used up to 4 times.

I then use Selective Color Correction tool to deal with remaining color issues.

You can use the SRDx tool if you are careful. LSI has a SRDx Photoshop Plug-in that is more useful. You can run it many times at different strengths on separate layers and use masks.

In Preferences Auto Tab uncheck the box Active for the Pipet tool. I also change the Highlight Offset to 6, and the Shadow Offset to 92.
This will make ACC set a white point at about 242 to 245. I save the exposure increase for LR as it does it intelligently.
The White Pipet will use the same setting. The Black Pipet will set R20, G20, B20. This is darker than I want. I only use it on photos where I can't find anything black and then guess at a black spot. Then set a color sampler and increase the numbers to mid 20's using Histogram Master Tab.
I rarely use the Black Pipet and you can change the preference to get a higher number.

hope this is useful.

[saiguy]

I don't try to match a print. The print may have issues from aging. Just try to get the best scan.

In Preferences Color Management, you can set the color space you want the file to be tagged with. Such as Adobe RGB.
Your iT8 profile should be there for Input. There is a pref I think in AUTO where you can check a box for Custom Naming of the profile. Such as the date created. I think by default is is off.

[eriksatie]

I don't try to match a print. The print may have issues from aging. Just try to get the best scan.

In Preferences Color Management, you can set the color space you want the file to be tagged with. Such as Adobe RGB.
Your iT8 profile should be there for Input. There is a pref I think in AUTO where you can check a box for Custom Naming of the profile. Such as the date created. I think by default is is off.

Thanks saiguy, I've done all of that. About the fact to try to match the print: it is a workflow issue. I'm scanning while working in my office on other tasks so I'm trying to get a "neutral" scan to work on later. I have thousands to scan!

With negatives I've used a different approach: I used Negafix and Global CC in Silverfast to have a good starting point to work on later with no obvious color casts. But with prints and slides I was hoping to do less work in Silverfast. Just scan with custom ICC profile and that's it, no needs to correct colorcast. Match the print is the easiest approach for me to get the scan as neutral as possible.

[eriksatie]

I think I'm on something: the colorcast seems to be caused be the normalization of the histogram in Silverfast. Just doing a flat scan, with no normalization, seems to produce a better match with the print.

This surprise me a lot. I was thinking that Histogram normalization wouldn't affect colors but in effect it does. looking at LAB values L changes (and this was expected) but also A and B channels. It has the effect of exaggerate the dominant cast. A and B channel values goes further away from zero.

I've done the same test in Affinity Photo with the same result.

Someone can explain me why?

as reference, I'm attaching the last scan with Histogram normalization that matches quite closely the print

[eriksatie]

My workflow in SF8. Give this a try. I find the Global CC too difficult to work with.

Pre scan and set the frame. Hit the ACC, auto color correction. This will set a very good white point unless it keys on some specular highlight. If that happens I will increase the brightness with the top slider in Picture Settings. For sun set shots the Landscape choice seems to work best.

Open the Histogram tool and use the Separate tab. Undock the tool and you can drag the bottom corner to stretch it across the screen. Then for each RGB channel set a good black point by moving the left slider just a little inside where the data starts.

Open the Densitometer tool. By shift clicking you can set and remove colors samplers, up to 4. Try to find what should be the darkest black place in the photo. If the RGB numbers are very close you have a neutral black point. If not, undock the Histogram again and you can fine tune the numbers on the sampler.

If you see a color cast, the Neutral Pipet tool can remove it if you find the right place to click. But you need to set in Preferences for the tool to be used for that, otherwise the place you click will go R128, G128, B128. Some times I set a color sampler before clicking with the Neutral Pipet. This can be used up to 4 times.

I then use Selective Color Correction tool to deal with remaining color issues.

You can use the SRDx tool if you are careful. LSI has a SRDx Photoshop Plug-in that is more useful. You can run it many times at different strengths on separate layers and use masks.

In Preferences Auto Tab uncheck the box Active for the Pipet tool. I also change the Highlight Offset to 6, and the Shadow Offset to 92.
This will make ACC set a white point at about 242 to 245. I save the exposure increase for LR as it does it intelligently.
The White Pipet will use the same setting. The Black Pipet will set R20, G20, B20. This is darker than I want. I only use it on photos where I can't find anything black and then guess at a black spot. Then set a color sampler and increase the numbers to mid 20's using Histogram Master Tab.
I rarely use the Black Pipet and you can change the preference to get a higher number.

hope this is useful.

Thanks for this, it's different from my approach (I don't use ACC usually) but I'll try. With strong color cast on color negatives I find Global CC very effective and fast. Also Neutral Pipet is very effective, if you can find a neutral grey spot (not always the case)

[degrub]

perhaps the images are under exposed ?

[saiguy]

If you are using SF8 Ai you could scan in HDR mode and do the corrections in HDR Studio.
If you do not have HDR Studio, you can buy it for $50 USD. At least that's what I did some years ago. The difference in price between SF8 Ai Studio and The Archive Suite, with includes HDR, is about 50 dollars. You should call LSI sales if this appeals to you.

ACC does a great job with the Histogram except for the black point. It is way different than pulling the highlight sliders in. Sometimes they will be very far to the right of the data. But the image will look good.

[digitaldog]

Another test (same paper)
The first image has no correction, just the custom profile applied and then converted in Adobe RGB

And it matches the print? IF so, the profile is sound. The profile doesn't correct anything and IF the original is in need of further edits, that's why you have other controls in the scanning GUI to 'fix' that.

[eriksatie]

And it matches the print? IF so, the profile is sound. The profile doesn't correct anything and IF the original is in need of further edits, that's why you have other controls in the scanning GUI to 'fix' that.

no, it doesn't much very well the print. but please look a bit lower, there is another post about histogram normalization that could have been the source of the problem

[eriksatie]

If you are using SF8 Ai you could scan in HDR mode and do the corrections in HDR Studio.
If you do not have HDR Studio, you can buy it for $50 USD. At least that's what I did some years ago. The difference in price between SF8 Ai Studio and The Archive Suite, with includes HDR, is about 50 dollars. You should call LSI sales if this appeals to you.

ACC does a great job with the Histogram except for the black point. It is way different than pulling the highlight sliders in. Sometimes they will be very far to the right of the data. But the image will look good.

I know about HDR and I'm tempted, would be useful for the amount of scan I have to do, but now the upgrade is 200€.. costly
ACC is quite hit or miss for me, I haven't used much but it seems to work for slight casts, no use if you have a strong cast due to film aging

[Doug Gray]

Flat bed print scanners are simply not colorimetrically accurate. One well known problem is that scanners use RGB filters much like a camera and the combination of their illumination type (CFL or LED) and RGB filters do not meet Luther/Ives criteria. This can be dealt with by profiling a scanner against a profile generated from a patch set printed by the same printer on the same paper as was used for printing the photos to be scanned.

This is a tedious process and not offered by any company I'm aware of. However, it can be done using Argyll's s/w.

But there's another problem that can be bigger. It's large area crosstalk. Turns out scanners illuminate the scanned image in such a way that the image is illuminated in part by light reflected off nearby portions of the document. This can result in as much as a 20% difference in the light reflected from the print and can alter the scanned L* (and a*, b*) . An area of scanned print that is exactly the same as another area can see large differences in the Lab values that should be the same. This is not correctable in any sense by even the best profiles. It's an intrinsic, non-linear, property of essentially all flat bed scanners. This doesn't show up on standard scanner profile reports because they are self referring. To see it you have to scan targets where the patch colors are scrambled around to different locations than on the one used to create the profile. I ran into this after trying to create good profiles with thousands of patches and seeing significant dE variation when scanning patches printed in different locations.

I was shocked at how large the variations could be and discussed it here:

https://forum.luminous-landscape.com/index.php?topic=125738.msg1054468#msg1054468

To sum up, don't expect scans to be colorimetric representations of what is scanned. At best they can look reasonably good and w/o an overall cast.

[eriksatie]

Flat bed print scanners are simply not colorimetrically accurate. One well known problem is that scanners use RGB filters much like a camera and the combination of their illumination type (CFL or LED) and RGB filters do not meet Luther/Ives criteria. This can be dealt with by profiling a scanner against a profile generated from a patch set printed by the same printer on the same paper as was used for printing the photos to be scanned.

this, if I understand correctly, seems to confirm my impression that prints made with the same paper of the IT8 target scan better then prints made with other kind of paper

Flat bed print scanners are simply not colorimetrically accurate. One well known problem is that scanners use RGB filters much like a camera and the combination of their illumination type (CFL or LED) and RGB filters do not meet Luther/Ives criteria. This can be dealt with by profiling a scanner against a profile generated from a patch set printed by the same printer on the same paper as was used for printing the photos to be scanned.

This is a tedious process and not offered by any company I'm aware of. However, it can be done using Argyll's s/w.

But there's another problem that can be bigger. It's large area crosstalk. Turns out scanners illuminate the scanned image in such a way that the image is illuminated in part by light reflected off nearby portions of the document. This can result in as much as a 20% difference in the light reflected from the print and can alter the scanned L* (and a*, b*) . An area of scanned print that is exactly the same as another area can see large differences in the Lab values that should be the same. This is not correctable in any sense by even the best profiles. It's an intrinsic, non-linear, property of essentially all flat bed scanners. This doesn't show up on standard scanner profile reports because they are self referring. To see it you have to scan targets where the patch colors are scrambled around to different locations than on the one used to create the profile. I ran into this after trying to create good profiles with thousands of patches and seeing significant dE variation when scanning patches printed in different locations.

I was shocked at how large the variations could be and discussed it here:

https://forum.luminous-landscape.com/index.php?topic=125738.msg1054468#msg1054468

To sum up, don't expect scans to be colorimetric representations of what is scanned. At best they can look reasonably good and w/o an overall cast.

Thanks for this general explanation. yes, I was expecting that a scanner just measure the colors of each point of a media you put to scan, with an error that could be (almost) completely compensated by a custom profile. but it seems more difficult than this. I start to think that profiling is not that useful... with all these variables going, what's the point to profile at all?

I'm gonna read the topic you link, being about the same scanner and similar profiling problem, albeit a bit technical for me : )

[saiguy]

I just looked at LSI web site and it appears that HDR is $200 more now. But if you call LSI sales and tell them you already have a SF8 license you might get a pleasant surprise. Worth a try. From HDR you can out put different versions from the same scan and compare. Plus the scanner isn't running while you are doing the develop part. And you get a RAW archive file.

No reason not to calibrate with iT8. It's a one clicker. "with all the variables" a non calibrated profile will give you another one with less reliable color rendition. Yes, lots of variables. That's the challenge of scanning. Best practice is to get the best starting scan. It requires some work and skill.

[eriksatie]

I just looked at LSI web site and it appears that HDR is $200 more now. But if you call LSI sales and tell them you already have a SF8 license you might get a pleasant surprise. Worth a try. From HDR you can out put different versions from the same scan and compare. Plus the scanner isn't running while you are doing the develop part. And you get a RAW archive file.

I'm evaluating HDR because it would me allow to scan during the day (while working) and edit the scan in the evening, saving me lots of time during the day. It seems to work well, Auto CCR seems to work better then in the Studio. on the same scan it does not apply the same exact settings as Studio, perhaps because it is working on the full scan instead of a preview.

I've called Silvefast Europe, because I live in Italy, but unfortunely they doesn't offer me a discount : ( the upgrade from Silverfast Studio it's 200€ but I think I will buy it anyway because I will really save a lot of time